HYBRID DEEP LEARNING MODEL FOR UPI FRAUD DETECTION USING CNN AND RANDOM FOREST

With the increasing adoption of Unified Payments Interface (UPI) transactions, fraud has surged, posing a significant risk to users and financial institutions. Traditional fraud detection methods often struggle to keep pace with evolving attack patterns, leading to false positives and missed fraudulent cases. To address this, we propose a hybrid deep learning model that combines Convolutional Neural Networks (CNN) and Random Forest (RF) to improve fraud-detection accuracy. In this study, CNN was used to extract high-level transactional patterns and capture complex relationships within the financial data. The extracted features are then classified using RF, which enhances interpretability and robustness. The dataset used consists of real-world UPI transactions preprocessed using standard scaling and noise reduction techniques to improve generalization. The proposed model was trained and evaluated using a balanced dataset to mitigate class imbalance issues. The experimental results demonstrate that the proposed CNN-RF model achieves high accuracy while maintaining a low false-positive rate. Compared to standalone models, this hybrid approach significantly enhances fraud-detection efficiency. This study highlights the importance of combining deep learning with traditional machine learning techniques for real-time fraud detection in digital payments. Future work will include the integration of real-time anomaly detection and adaptive learning techniques to further enhance security.